Using a massively scaled-up version of "Proximal Policy Optimization", which is a training system designed for OpenAI's successful 1v1 Dota 2 bot. OpenAI Five plays the game for a whopping 180 years every single day thanks to 256 GPUs and 128,000 CPU cores in the Google Compute Cloud.
The humans did eventually manage to grab game three from the bots but the damage was done.
After OpenAI took down members of the audience easily during a warmup round, the bot network started and finished the first of the two matches strongly, stopping its human opponents from destroying any of its defensive towers. "This reaction time is much closer to human level, though we haven't seen evidence of changes in gameplay as OpenAI Five's strength comes more from teamwork and coordination than reflexes". If the the OpenAI Five are victorious, they not only join the pantheon of machines that have publicly beaten humans at games (Deep Blue, Watson, AlphaGo), but will have done so in a game that balances several real-time interactions at once.
OpenAI Five has been working towards their goal of competing against the best pro team at The International which is the largest annual Dota tournament in the world. "You have imperfect information, you have team work, you have these exponential combinations of different heroes and items, and you have to be able to deal with all of that". It won the first game in 21 minutes and 37 seconds.
In preparation this year's Valve International 8 later this month has an OpenAI Five Benchmark match was held in San Francisco on August 5th.
The tournament was a best of three but clearly designed for fun more than goal. Valve's multiplayer online battle arena (MOBA) video game, Dota 2, is the latest example of that. In the third match, the Twitch chat audience chose the worst combination of heroes for the bots to play and the A.I. predicted a 2.9% chance to win. The Open AI bot went on to best other notable players including, SumaiL, and former world champion, Dendi.